Re: [PATCH] drivers/perf: riscv: handle legacy fallback in event check

[Will Deacon]

On Fri, Jul 10, 2026 at 01:08:35PM +0300, Sergey Matyukevich wrote:
> Common events PERF_COUNT_HW_CPU_CYCLES and PERF_COUNT_HW_INSTRUCTIONS
> can fall back to the legacy counters. So do not mark them as invalid
> if OpenSBI reports that they are not mapped to sampling counters
> in PMU DT node.
> 
> Signed-off-by: Sergey Matyukevich <geomatsi@xxxxxxxxx>
> ---
>  drivers/perf/riscv_pmu_sbi.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/perf/riscv_pmu_sbi.c b/drivers/perf/riscv_pmu_sbi.c
> index dfc886dee5ad..8f7aa71dc96f 100644
> --- a/drivers/perf/riscv_pmu_sbi.c
> +++ b/drivers/perf/riscv_pmu_sbi.c
> @@ -335,7 +335,8 @@ static int pmu_sbi_check_event_info(void)
>  		goto free_mem;
>  	}
>  
> -	for (i = 0; i < ARRAY_SIZE(pmu_hw_event_map); i++) {
> +	/* skip check for cycles and instructions as they can fall back to legacy counters */
> +	for (i = 2; i < ARRAY_SIZE(pmu_hw_event_map); i++) {
>  		if (!(event_info_shmem[i].output & RISCV_PMU_EVENT_INFO_OUTPUT_MASK))
>  			pmu_hw_event_map[i].event_idx = -ENOENT;
>  	}

I'm assuming the riscv tree are handling perf patches to the riscv driver,
but please shout if not.

Will